BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Towards Graph-based Diffusion Models in Digital Pathology
DTSTART:20230620T100000
DTEND:20230620T120000
DTSTAMP:20260407T075544Z
UID:07ffd27be6b6b29248baba7adf1a5e3976869a07ff58a33c03796a7e
CATEGORIES:Conferences - Seminars
DESCRIPTION:Manuel Monteiro Lança Madeira \nEDIC candidacy exam\nExam pr
 esident: Prof. Pierre Vandergheynst\nThesis advisor: Prof. Pascal Frossard
 \nThesis co-advisor: Dr. Dorina Thanou\nCo-examiner: Prof. Maria Brbic\n\n
 Abstract\nGraph deep learning approaches to digital pathology data have be
 en particularly successful due to their capability of capturing complex de
 pendencies between tissue entities\, such as cells. The usage of deep gene
 rative models in this setting holds great promise by enabling the augmenta
 tion of the scarce digital pathology data and providing a more interpretab
 le framework for the analysis of the reproduced biological mechanisms.\nMo
 tivated by these premises\, we first analyse a paper that leverages Graph 
 Neural Networks to capture and characterize disease-relevant microenvironm
 ents\, showcasing the advantages of such techniques to digital pathology. 
 We proceed to the introduction of diffusion models through the analysis of
  the paper that laid its empirical foundations as a state-of-the-art appro
 ach in generative modelling. Then\, we consider the adaptation of diffusio
 n methods to discrete state-spaces\, setting a first step towards the unif
 ication of the two former papers\, i.e.\, graph- based diffusion models. F
 inally\, we briefly discuss open research directions that promise to furth
 er improve the generation of interpretable and biologically plausible data
 \, as the incorporation of biological priors and hierarchical generation s
 chemes.\n\nBackground papers\n\n	Denoising Diffusion Probabilistic Models 
 (https://proceedings.neurips.cc/paper_files/paper/2020/file/4c5bcfec8584af
 0d967f1ab10179ca4b-Paper.pdf)\n	Structured Denoising Diffusion Models in D
 iscrete State-Spaces (https://proceedings.neurips.cc/paper_files/paper/202
 1/file/958c530554f78bcd8e97125b70e6973d-Paper.pdf)\n	Graph deep learning f
 or the characterization of tumour microenvironments from spatial protein p
 rofiles in tissue specimens (https://www.nature.com/articles/s41551-022-00
 951-w)\n
LOCATION:ELE 242 https://plan.epfl.ch/?room==ELE%20242
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR